I'm glad you found the discussion helpful! To be clear, my concerns aren't about synthetic data, but about synthetic text, i.e. setting up systems where lawyer (etc) take the output of an LLM as if it were information.
LLMS
-
Satirical take on RAG papers and retrieval prompting
By
–
every RAG paper starts like We achieve SoTA performance on this challenging benchmark via a retrieval prompting pipeline paired with a synthetically generated knowledge base of interpreter-verified solutions, a method we call RAG Against the Machine.
-
Grok Mini vs Full Grok: When to Use Each Model
By
–
Grok Mini is best for most queries. Full Grok is best used for complex questions.
-
Optimizing AI Inference Speed for Faster Response Times
By
–
About a week or so. We need to optimize the inference speed, so answers appear faster.
-
Exploring GPTs for Marketing Strategy and SEO
By
–
Explore GPTs in Marketing Strategy Boosts content creation, SEO • Generate engaging prompts
• Optimize for search engines
• Target relevant audiences Read more: https://
buff.ly/3SHJdTF -
Abliterating Models Without LoRA Adapters Techniques
By
–
You'd need to abliterate it without LoRA adapters like in this notebook:
-
LazyMergekit: Merge AI Models with Commands and Config
By
–
You can do it with LazyMergekit. I've added all the commands and the config I've used.
-
Removing AI Censorship: Standardized System Prompt Approaches
By
–
That's what I thought but I didn't manage to remove the censorship. Is there a standardized system prompt I should use for that?
-
GGUF Quantizations Ready for Hermes-3-Llama Model Testing
By
–
GGUF quants are also ready if you want to test the model. @bartowski1182 already made fancy imatrix ones. Love the new model tree on @huggingface
! (cc @victormustar @julien_c
) GGUF: https://
huggingface.co/mlabonne/Herme
s-3-Llama-3.1-8B-lorablated-GGUF
… -
Open-Source AI Models and Abliteration Techniques Explained
By
–
Of course, special thanks to @NousResearch and @teknium for these high-quality models. Thanks to grimjim and @failspy for the abliteration techniques that were used here. You can learn more about it in my article: